M
Persistent memory
Agents remember decisions, patterns, fixes, and project context across sessions — no more re-explaining the same architecture on day two.
Self-hosted agent memory
Your AI agents finally share the same brain.
Claude, Codex, Cursor, Gemini, and Aider stop losing context between sessions. One shared memory layer, running on your own machine, that every agent you use can read and write.
Available for Windows, macOS, and Linux. Community is free with no time limit; the desktop app starts its embedded services automatically.
agent-brain://memorylocal
brain({
command: "remember",
input: {
type: "decision",
project: "checkout",
content: "Use Stripe webhooks for fulfillment",
},
})
brain({
command: "recall",
input: { query: "how do we ship purchases?" },
})
// shared across Claude, Codex, Cursor, or REST40-70%
token reduction from semantic cache
57ms
P95 recall latency in benchmarks
17
MCP commands for shared memory
Free
community tier, self-hosted
See it in action
Watch the memory layer in action.
Agent Software Suite workflow
See voice capture, project memory, and bounded task delegation working together in one session.
What it does
Context loss between sessions is the friction. This is what removes it.
Agent Brain runs a shared memory store, semantic cache, and governance layer on your own machine — one context layer that every connected agent reads and writes, instead of each tool starting over from zero.
C
Semantic cache
Similar prompts reuse prior answers instead of burning tokens on a question that's already been asked and answered.
N
Cross-agent context
Claude, Codex, Cursor, Gemini, Aider, and custom agents all read and write the same memory — no more siloed context per tool.
L
Runs on your own hardware
Memory storage and recall run in local services managed by the desktop app. Your project context stays on infrastructure you control.
A
Memory governance
Audit, deduplicate, archive, restore, and consolidate memories so recall stays sharp as the corpus grows instead of degrading.
A
Apply audit fixes
Review proposed memory changes, then apply approved audit findings in one controlled step instead of repairing the corpus by hand.
D
Desktop control plane
One desktop app handles setup, health checks, knowledge exploration, and monitoring for the whole stack.
How it works
Local services, one shared context layer.
1
Start the local stack
The desktop app boots the local services Agent Brain needs — no cloud account to provision first.
2
Connect your agents
Point Claude, Codex, Cursor, Gemini, or Aider at the local MCP endpoint and they're wired in.
3
Store durable knowledge
Agents write decisions, fixes, and context as structured, scoped memories instead of losing them at session end.
4
Recall what matters
The next session — in any connected tool — retrieves the relevant context automatically.
Local memory store
Fast recall layer
Shared agent context
Governance workflows
Local model support
MCP + REST adapters
Adapter layer
MCP clients use localhost:3100/mcp. REST and WebSocket clients use localhost:9090/api/v1.
Comparison
Built for coding-agent workflows.
Self-hosted
Agent BrainYesmem0NoGPTCacheNoLettaNo
Semantic cache
Agent BrainYesmem0NoGPTCacheYesLettaNo
Cross-agent sharing
Agent BrainYesmem0NoGPTCacheNoLettaNo
Local model operations
Agent BrainYesmem0NoGPTCacheNoLettaNo
Desktop stack control
Agent BrainYesmem0NoGPTCacheNoLettaYes
MCP support
Agent BrainYesmem0YesGPTCacheNoLettaNo
Agent-agnostic adapters
Agent BrainYesmem0NoGPTCacheNoLettaNo
Community tier
Agent BrainYesmem0NoGPTCacheYesLettaYes
| Capability | Agent Brain | mem0 | GPTCache | Letta |
|---|---|---|---|---|
| Self-hosted | Yes | No | No | No |
| Semantic cache | Yes | No | Yes | No |
| Cross-agent sharing | Yes | No | No | No |
| Local model operations | Yes | No | No | No |
| Desktop stack control | Yes | No | No | Yes |
| MCP support | Yes | Yes | No | No |
| Agent-agnostic adapters | Yes | No | No | No |
| Community tier | Yes | No | Yes | Yes |
Pricing
Start free. Upgrade when the memory cap gets in the way.
Community runs the full stack with real limits. Pro removes them — unlimited memories, semantic cache, and desktop backup — for commercial work.
FAQ
Common questions
Does Agent Brain send my code or memories to a cloud service?
No. Agent Brain is self-hosted. Core memory and recall operations run on your machine or infrastructure.
Which agents can use it?
Claude, Codex, Cursor, Gemini, and Aider connect out of the box through MCP. Custom agents and scripts can connect through the same REST interface.
What does the semantic cache do?
It checks whether a similar query already has a useful answer. If the similarity score clears your threshold, agents can reuse the cached response instead of spending fresh tokens.
Do I need Docker?
No. The desktop app starts Agent Brain's embedded services automatically. Docker is optional for developers who choose to run the source or Compose stack directly.
Is Pro checkout live?
Yes. Click 'Get Pro' on the pricing page to subscribe monthly or annually. Community downloads are available now from the download page.
Can I use Community for commercial work?
No — Community is for personal and non-commercial use only. If you are using Agent Brain as part of paid work, a business, or a team, you need a Pro or Enterprise license.
Affiliate Program
Promote Agent Software Suite. Earn 30% recurring.
Refer paying builders and earn 30% on every invoice they pay for the next 12 months.
30%
Commission
12 invoices
Window
60 days
Cookie
Stop re-explaining your codebase every session.
Launch the local services, point Claude, Codex, Cursor, Gemini, or Aider at them, and let every agent share the same memory.